Table 2.

Summary of segmentation performance measures (means ± stds) of the two cartilage classes of our proposed methods DMA-GCN (CT/SEQ) by varying the overall learning paradigm (transductive vs. inductive). Best results for each category (CT vs SEQ) with respect to DSC index are highlighted.

			Femoral Cartilage					Tibial Cartilage
Module	SEQ	CT	Recall	Precision	$\mathcal{DSC}$	$\mathcal{VOE}$	$\mathcal{VD}$	Recall	Precision	$\mathcal{DSC}$	$\mathcal{VOE}$	$\mathcal{VD}$
Inductive	✓		$91.78\%$	$89.61\%$	$\mathbf{89.45}\%$	$15.11\%$	$5.98\%$	$86.09\%$	$84.29\%$	$83.81\%$	$26.02\%$	$8.45\%$
			$(\pm 0.051)$	$(\pm 0.038)$	$(\pm 0.086)$	$(\pm 0.093)$	$(\pm 0.067)$	$(\pm 0.036)$	$(\pm 0.053)$	$(\pm 0.041)$	$(\pm 0.094)$	$(\pm 0.078)$
		✓	$85.78\%$	$87.71\%$	$85.87\%$	$23.61\%$	$7.79\%$	$85.92\%$	$84.93\%$	$\mathbf{84.05}\%$	$26.15\%$	$8.71\%$
			$(\pm 0.045)$	$(\pm 0.033)$	$(\pm 0.056)$	$(\pm 0.069)$	$(\pm 0.041)$	$(\pm 0.049)$	$(\pm 0.071)$	$(\pm 0.046)$	$(\pm 0.029)$	$(\pm 0.027)$
Transductive	✓		$96.13\%$	$95.21\%$	$95.44\%$	$13.21\%$	$3.98\%$	$95.19\%$	$94.41\%$	$93.87\%$	$18.65\%$	$5.03\%$
			$(\pm 0.121)$	$(\pm 0.183)$	$(\pm 0.022)$	$(\pm 0.045)$	$(\pm 0.089)$	$(\pm 0.031)$	$(\pm 0.027)$	$(\pm 0.024)$	$(\pm 0.034)$	$(\pm 0.023)$
		✓	$96.17\%$	$95.81\%$	$\mathbf{95.71}\%$	$13.17\%$	$3.94\%$	$95.31\%$	$94.78\%$	$\mathbf{94.02}\%$	$17.98\%$	$4.99\%$
			$(\pm 0.117)$	$(\pm 0.189)$	$(\pm 0.032)$	$(\pm 0.051)$	$(\pm 0.094)$	$(\pm 0.032)$	$(\pm 0.025)$	$(\pm 0.021)$	$(\pm 0.034)$	$(\pm 0.029)$